Ooooooh John Bongaarts seems to have an interesting argument here. How long do we live, not as long as we think we do. This paper is hellishly technical, and I don't doubt controversial, but the argument seems sound. There is a bias in the way we calculate period life expectancies and age-specif death rates. These biases Bongaarts refers to as mortality tempo bias. To try and explain this simply it is really symmetrical with the easier to understand tempo distortions in fertility. The fertility rate seems lower if measured at a time when first conception age is rising. By the same measure, life expectancy seems longer, when measured during the upward curves of expectancy, because in given years less people will die. What does this mean for the accuracy of our expectancy and social security calculations. Not a lot when the median death age is only rising slowly. This potential upward bias is as nothing when compared to the potential downside biases which are normally incorporated in the forward looking life expectancy projections. It all depends on what changes take place in the rate at which life expectancy increases. It should be noted that most of the calculations assume a return to population reproducing fertility rates, and a reduction in the rate of increase in life expectancy with very little theoretical justification whatever. we just hope it's going to be like this. (I use hope advisedly here, since I am talking of the expectations of people who run pension funds, not our individual 'hope' to live longer. Perhaps this is an unfortunate use of the verb 'hope').
The principles underlying methods for the study of mortality have tended to be accepted without question. In particular, age-specific death rates and life table measures
of mortality calculated from them are widely regarded as beyond methodological reproach. We have argued that for countries with high life expectancies, in which nearly all mortality occurs beyond young adulthood, period age-specific death rates are subject to a significant bias whenever these rates are changing. When the mean age at death rises, the rates are biased downward and when the mean age at death falls, they are biased upward. Following related work on the measurement of fertility, we refer to these as mortality tempo biases.
Period life expectancies calculated from age-specific death rates by life table methods are biased as well, and we have provided a formula for determining the magnitude of the bias. The estimated average tempo effect in conventional life expectancy for females ranges from 1.6 years in the US and Sweden to 3.3 years in Japan for the period 1980–95.
Our main finding is that the conventional calculation of period life expectancy at birth gives a misleading indication of how long we live. We are not living as long as we thought we were. This result is significant in its own right, but it also has implications for the study of past and future trends in life expectancy, because tempo effects vary over time.
Source: How Long Do We Live? John Bongaarts, Griffith Feeney